AOS Community Edition — Open Source Agent Operating System

Introduction

AOS Community Edition is an open-source operating system purpose-built for AI agents. Written in Rust, it provides the scheduling, isolation, resource management, and inter-agent communication primitives that autonomous agents need to run reliably in production. Think of it as a lightweight OS kernel where the processes are agents, not traditional programs.

What AOS CE Does

• Schedules and manages the lifecycle of multiple concurrent AI agents
• Provides sandboxed execution environments with configurable resource limits
• Handles inter-agent message passing and coordination through typed channels
• Manages tool access and permissions so agents only reach resources they are authorized to use
• Logs every agent action for observability and audit compliance

Architecture Overview

AOS CE is structured as a microkernel written in Rust. The kernel handles agent scheduling, memory allocation, and IPC. Each agent runs in its own lightweight sandbox with a capability-based permission model — an agent must hold a capability token to access a tool, file, or network endpoint. The system supports both local single-machine deployments and distributed multi-node clusters where agents can migrate between hosts. A TOML-based manifest defines each agent's configuration, dependencies, and resource limits.

Self-Hosting & Configuration

• Download the pre-built binary or compile from source with Cargo
• Define agents in TOML manifest files specifying model, tools, and resource caps
• Configure the runtime to limit CPU time, memory, and network access per agent
• Set up the built-in telemetry exporter to send metrics to Prometheus or OpenTelemetry
• Enable cluster mode by pointing multiple AOS instances at a shared coordination backend

Key Features

• Capability-based security model prevents unauthorized tool access by agents
• Lightweight Rust runtime with minimal overhead per agent
• Typed IPC channels for safe inter-agent communication
• Built-in resource quotas prevent runaway agents from consuming all system resources
• Full action logging for compliance and debugging

Comparison with Similar Tools

• Docker / Containers — isolates processes; AOS CE isolates agents with agent-specific primitives
• Kubernetes — orchestrates services; AOS CE orchestrates autonomous agents with tool permissions
• CrewAI — Python multi-agent framework; AOS CE operates at the OS level with Rust performance
• AutoGen — conversation-based agents; AOS CE provides lower-level scheduling and isolation

FAQ

Q: What language are agents written in? A: Agents can be written in any language. AOS CE manages them as processes or WASM modules.

Q: Can agents communicate with each other? A: Yes. AOS CE provides typed IPC channels and a publish-subscribe message bus.

Q: Does AOS CE require a GPU? A: No. The runtime itself is CPU-only. Agents can access GPUs through configured tool capabilities.

Q: Is there a commercial edition? A: The Community Edition is fully open source. Enterprise features are planned for a separate edition.

Sources

• https://github.com/unicity-aos/aos-ce